[R] cpquery problem

[Marco Scutari]

Hi Ross,

On 31 July 2016 at 09:11, Ross Chapman <ross.chapman at ecogeonomix.com> wrote:
> I have tried running the cpquery in the debug mode, and found that it
> typically returns the following for instances where the conditional
> probability is returned as 0:
>
>    > event matches 0 samples out of 0 (p = 0)
>
> Am I right in understanding that the Monte Carlo sampling has been unable
> to create any cases that match the query?  If so, why would this be if the
> evidence used is very typical of an average case in the data used to train
> the network?

Yes, that is what is happening. As to the reason why, I guess that the
dependencies in the data may not be adequately represented in the
fitted Bayesian network for some reason. What is apparent is that
(EST=='y' & TR>9 & BU>15819 &  RF>2989) has an associated probability
low enough that you do not observe any such sample in rejection
sampling. Now, that being the case, you have two options:

1) use a much larger "n" with a smaller "batch = 10^6" to generate a
lot more particles;
2) switch to likelihood weighting, i.e.

cpquery(fitted,event=(ABW<=11), evidence=list(EST ='y', TR = c(9,
max(data$TR)), BU = c(15819, max(data$BU)), RF = c(2989,
max(data$RF)), n=10^6, method = "lw")

3) look at the parameters in your fitted network and diagnose why this
is happening.

Cheers,
    Marco

-- 
Marco Scutari, Ph.D.
Lecturer in Statistics, Department of Statistics
University of Oxford, United Kingdom